The association between early life mental health and alcohol use behaviours in adulthood: A systematic review

This systematic review aims to summarise current evidence on the association between early life mental health and alcohol use behaviours in adulthood. Peer-reviewed publications were located by searching EMBASE, Medline, PsycINFO, and the ISI Web of Science up to 31 October 2018. Prospective longitudinal studies reporting associations between externalising problems (EXT), internalising problems (INT), depression, anxiety before age 18, and alcohol use behaviours (alcohol consumption, heavy/problematic drinking, alcohol use disorder) after age 18 were included. After screening 17259 articles, 36 articles met the inclusion criteria. Information extracted included strength of associations, age when mental health and alcohol use behaviours were measured, sex differences in the association, and other sample characteristics. 103 tests in 23 articles were identified on the externalising domain and 135 tests in 26 articles on the internalising domain. 37 out of 103 tests reported positive associations between EXT and alcohol use behaviours. The likelihood of observing positive associations was higher for more severe alcohol use outcomes, but this trend disappeared among high-quality studies. Findings on associations between internalising domain and alcohol use varied across their subtypes. INT tended to be negatively associated with alcohol consumption but positively associated with more severe outcomes (heavy/problematic drinking, alcohol use disorder). Depression tended to be positively associated with alcohol outcomes, while no clear association between anxiety and alcohol outcomes was evident. Variation of the association across developmental timing, sex, culture, historical period was explored where appropriate. Great heterogeneity in the current literature calls for greater attention to view the relationship developmentally.


Introduction
Alcohol use is a major public concern, being responsible for 3.8% of deaths worldwide in 2004 with the proportion increasing to 5.9% in 2012. The inestimable burden suffered by individuals, families, and society due to alcohol-related problems has also compelled scholars' attention [1][2][3]. Identifying modifiable risk factors and the interactions among them is key to successful Exposure. Mental health problems were categorised into externalising problems (EXT), internalising problems (INT), depression, and anxiety. Under the externalising domain, we focused on general measures of EXT and conduct problems and did not include attention deficit hyperactivity disorder which is under the externalising domain but does not contain features that contribute to the externalising pathway [13]. We excluded studies measuring specific symptoms or traits, such as stealing or fighting. Studies with a wide age range population over age 18 were included only if the upper age boundary (two standard deviations above the mean age) was below age 18 to ensure mental health problems were measured below age 18 for the majority of the population. Studies that derived trajectories for mental health problems beyond age 18 were included only if the derived trajectories mainly reflected mental health status across childhood or adolescent (i.e., more than half of the measurement occasions occurred before age 18).
Outcome. We included all alcohol-specific outcomes and excluded substance use outcomes that did not explicitly represent alcohol use. For clarity, we further categorized alcohol use behaviours into three broad categories: alcohol consumption including drinking frequency/volumes; heavy/problematic drinking, including binge drinking, heavy drinking, and problematic drinking identified through well-known scales (e.g., Cut-down Annoyed Guilty Eye-open (CAGE) / Alcohol Use Disorder Identification Test (AUDIT)); AUD diagnosed based on the Diagnostic and Statistical Manual of Mental Disorders (DSM). As mentioned above, the included studies all measured alcohol use behaviour at or after age 18 as it is the minimum legal drinking age in most countries [44]; studies with a wide age range population under age 18 were included only if the lower age boundary (two standard deviations below the mean age) was at or above age 18. Studies that derived trajectories for alcohol use below age 18 were included only if the derived trajectories mainly reflected alcohol use in adulthood (i.e., more than half of the measurement occasions occurred after age 18). In addition, all studies included in this review had alcohol outcomes that were measured at least one year after the mental health measurements were taken to reflect the long-term prospective association between them.

Screening and data extraction
Guidelines set forth by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) were followed to ensure transparency [45], and the protocol for this systematic review was published on PROSPERO (registration number: CRD42018115502).
After excluding 5833 duplicates, 17,259 articles were screened for inclusion, and 15% of them were independently screened by DG. The agreement rate was 97.9% and Cohen's kappa was 0.71 at this stage. Disagreements were discussed and consensus was reached before screening the rest of the articles. After excluding 16768 articles based on the titles and abstracts, 495 articles were retrieved and assessed for eligibility. Ten percent of the full texts were screened by DG, and the agreement rate and Cohen's kappa at this stage were 92.3% and 0.75, respectively. The final sample constituted 36 articles, comprising 33 articles that met the eligibility criteria as well as three articles obtained through screening the references of eligible articles and relevant publications [16,28]. See more details in Fig 1 and Table 1.
An extraction form was developed by KN, and 10% of the selected articles were extracted by DG. The information extracted included author, year of publication, country (proxy for culture) and sampling strategy, sample size (proportion of male) and their birth year (proxy for history), measurement scale of exposure and outcome, age when exposure and outcome were measured, direction and size of the association, sex differences of the association, covariates adjusted for, statistical models, assessment of attrition bias, and methods for dealing with missing data. Associations were extracted if they reflected the total association of the relationship. For example, if depression at age 7 and depression at age 16 were adjusted for in the model simultaneously (e.g. outcome = a+b1 � depression at age 7+b2 � depression at age 16), then the coefficient b2 can be interpreted as total association between depression at age 16 and alcohol, which was not confounded by previous depression status (and thus was extracted), while the coefficient b1 was the controlled direct association between depression at age 7 and the outcome not through later depression status (and thus was not extracted). See more definitions in Pearl(2001) [79]. In addition, some studies reported several associations for the same exposure-outcome set, so other rules were devised to avoid duplicate associations: continuous measures (versus categorical), self-report (versus parent or teacher report), most properly adjusted result, unstandardized betas (versus standardised), whole population (versus subpopulation). Discrepancies were discussed and agreed upon before extracting information from all included articles. Early life mental health and adulthood alcohol use behaviours

Data synthesis
As shown previously [16], there was high heterogeneity in the studies included in this review in terms of sample characteristics (a wide age range of participants), subtype and developmental timing of mental health problems and alcohol use behaviours (binary/continuous, trajectory/one-time-point, measurement scales), length of follow-up, and confounders adjusted for. Approaches exist to overcome the heterogeneity due to analytical approaches [80], with the exception of exposure heterogeneity. Particularly, for continuous exposures, it was not possible to standardise or transform them to the same metric required by meta-analysis [81]. In addition, the number of articles is too low to be pooled after taking into account the potential factors examined in our study [82]. We, therefore, report results narratively, and present extracted associations in detail in S2 Table. To minimize potential bias caused by different ways of reporting the results (e.g. different exposure and outcome categories, reporting separately by sex/age), or by articles using the same population, data were synthesised in the following three ways: a) We reported the proportion of tests that were significant (P value equal or less than 0.05) out of all tests that reported specific exposure-outcome sets regardless of studies/articles as did by Hussong et al. [16]. , and neither of them was significant. Then, the proportion of negative association would be 1 / (4+2) = 16.7%. b) We reported the proportion of studies that reported significant association for each exposure-outcome set. For each study (using the same dataset) and for each exposure and outcome pair, no matter how many tests were reported, and the association was counted as significant as long as one test was significant. For instance, in the example above, article A would be counted as reporting a significant negative association between INT and alcohol consumption, and article B would be counted as reporting no association between INT and alcohol consumption. Then the proportion would be 1 / (1+1) = 50%. c) The method outlined in point a) was repeated in high-quality studies as defined below. We report results synthesised using the first method in the main article and other results in S3 and S4 Tables for readers' information. To maximize the use of available information and informed by the albatross plot [83], we also described the distribution of P-values for each association test against its sample size in Figs 2-6. In addition, average P-value across subtypes and developmental timing of both exposure and outcome was presented in S1-S4 Figs.

Quality assessment
We used an adapted version of the Critical Appraisal Skills Programme (CASP) Cohort Study Checklist [84], which was shortened to 8 questions, as shown in S5 Table. We mainly assessed four aspects of each cohort study: sample selection, measurement error, core confounder adjustment, and the handling of missing data, which are key issues that can cause bias in observational studies [85,86].
A quality score (QS) was assigned for each question. The scores were then summed, with total scores ranging from 0 to 8. Studies with scores ranging from 0-4 were considered as poor quality, and studies with score 5-8 as good quality. Quality assessment was done for all 36 selected articles by two researchers separately and disagreements were discussed to reach a final consensus.
We organise the results by four subtypes of mental health problems: EXT, INT, depression and anxiety. Within each domain, we further structure our findings for three subtypes of alcohol use behaviours. Where appropriate, we further explored whether the results were affected by whether adjusting for EXT and INT accordingly, the developmental timing of exposure and outcome, and country origin or birth cohort; we also summarised the evidence for potential sex differences.

Search results
Of the 36 articles included in this review, eleven studies were carried out in the US and nine in the UK, followed by six in Finland. The data used were from over 20 longitudinal studies, but six articles used data from the Avon Longitudinal Study of Parents and Children (ALSPAC). The sample size in 22 articles was over 1000.

Association between externalising problems and alcohol use behaviours
Our review identified 103 tests of the association between EXT and alcohol use behaviours in 23 articles, and higher early life EXT was significantly associated with more alcohol-related issues later in 37 tests (35.9%).
Then, we explored the variation of association according to the developmental timing excluding alcohol consumption outcomes. Two out of seven (28.6%) tests measuring EXT in childhood reported positive associations with later alcohol use behaviours [21,23,64], 10 out of 19 (52.6%) tests measuring EXT in early adolescence showed positive associations [21,24,47,51,52,56,63,71,74,77,78], and two tests measuring EXT in adolescence both reported positive associations [55]. Results from four papers using EXT trajectories indicated that EXT in adolescence might be more strongly related to alcohol outcomes, especially the persistence of EXT from childhood to adolescence [61,73,75,76] with the exception of Bor et al.'s study [9]. This association pattern was also reflected in papers that measured EXT at several time points [47,51,64] with the exception of Maggs et al.'s study [21]. Out of 41 tests measuring alcohol use in transition to adulthood, 23 (56.1%) presented positive associations between EXT and  Fig 3. Among the 23 articles, 19 included both males and females in their samples, and four explored associations only among males [23,[74][75][76]. Eleven of the 19 articles that included both sexes did not explore whether there was an interaction between sex and EXT in the association with later alcohol use [21,24,47,50,52,54,61,63,64,68,69]. Only two of the remaining articles reported a significant sex interaction with the association being stronger in males [56], while the other six articles found no statistically significant interaction [51,55,57,71,73,77]. Among tests reporting the association separately in male and female, 15 out of 42 (35.7%) in male were significantly positive; 6 out of 28 (21.4%) in female were significantly positive. To explore the role of culture and history, we categorized country origin into two groups (Europe versus non-Europe), and birth year into three cohorts (born in or before 1960s, born in 1970s, born in or after 1980s). Proportion of positive results were similar across continents (Europe Early life mental health and adulthood alcohol use behaviours 44.8% versus non-Europe 48.6%) (See Fig 3); four out of twelve tests (33.3%) among those born in or before 1960s reported positive associations, 15 out of 34 tests (44.1%) among those born in 1970s reported positive associations, and 11 out of 19 tests (57.9%) among those born in or after 1980s reported positive associations (See Fig 3). To tease out age effect from cohort effect, analysis was further limited to those born in 1970s and had their alcohol measured during transition to adulthood, 11 out 21 tests (52.4%) found positive results.
Summary. EXT was positively associated with later alcohol use and this association varies across subtypes of alcohol use behaviours: higher proportion of positive associations for more severe outcomes. More positive associations were detected when EXT was measured in adolescence and alcohol use in transition to adulthood. The probability of detecting significant positive associations between EXT and later alcohol use behaviours was higher when adjusting for INT simultaneously. Most of the studies that tested sex effect in the association detected no significant interaction, however, higher proportion of positive results were reported in male population. The probability of detecting a positive association between EXT and alcohol use behaviours appeared to be consistent across countries and cohorts.
Due to the divergent direction of the association, analysis was done with respect to each subtype of alcohol outcome and summarized as following: for alcohol consumption, higher proportion of negative associations was detected when EXT were simultaneously adjusted for (7 out of 8 tests vs. 5 out of 27 tests); for heavy/problematic drinking, it was more likely to detect positive associations when INT were measured at adolescence; for AUD, which was mainly measured during transition to adulthood, significant positive associations were reported when INT were measured during early adolescence and adolescence and when EXT were simultaneously adjusted for. There was no country-source heterogeneity within subtype of alcohol outcomes, but none of the studies using AUD as an outcome was from European countries. Cohort effect cannot be explored as the majority of the tests on alcohol consumption were from participants born in or before 1960s, and all tests on AUD were from participants born in 1970s.
No further exploration was carried out due to the limited number of tests for alcohol consumption and AUD. With respect to heavy/problematic drinking, among 12 tests that adjusted for EXT, four (33.3%) tests reported positive associations while two (16.7%) tests reported negative associations; among 12 tests that did not adjust for EXT, five (41.7%) found positive associations while one (8.3%) found negative associations. No conclusion can be drawn regarding the development period for depression as depression was mainly measured during adolescence. As for country and cohort differences, negative associations were only detected in one national study in the USA [48,49].
Anxiety and alcohol use. We identified 52 tests for the association between anxiety and alcohol use, one of which measured social anxiety [53]. Six out of 51 tests measuring general anxiety indicated positive associations [59,64] and five produced negative associations [72,74,78]. Negative association between social anxiety and alcohol consumption was reported [53].
For alcohol consumption, two out of thirteen tests in one article showed negative associations [72] and four tests found positive associations [59]. For heavy/problematic drinking, two out of 31 tests in two articles showed negative association [72,78], and no statistically significant association was detected for the remaining tests [72,77]. Out of seven tests identified from three articles with alcohol use disorder as an outcome [24,64,74], two tests found positive associations [64] and one test reported negative associations [74]. The distribution of P-value against sample size for anxiety and alcohol use behaviours is shown in Fig 6, and no systematic pattern of the association can be observed.
It should be noted that when anxiety was measured during early adolescence, only negative associations were found [72,74], while positive associations were only reported when anxiety was measured during adolescence [64,90]. No significant associations were reported when anxiety was measured during childhood. Two out of seven tests that adjusted for EXT reported negative associations, while three out of 44 tests found negative associations and six out of 44 tests reported positive associations when EXT were not adjusted for. No exploration for country or cohort effect can be done after taking into account the influence of developmental timing of anxiety.
Summary. Evidence for the association between internalising domain and alcohol use behaviours were inconsistent but somewhat varied across subtypes of the internalising domain and alcohol use behaviours. The relationship between INT and alcohol use behaviours tended to be negative for mild alcohol behaviour, especially when EXT was adjusted for, and positive for severe alcohol outcomes. The association between depression and alcohol outcomes seemed to be positive across subtypes. The association between anxiety and alcohol use behaviour was equivocal, and the reason might be that anxiety at different developmental timing was associated with later alcohol use behaviours in a different way.
24 out of 26 articles about the internalising domain had both males and females in their studies, and 11 of them did not explore sex differences in the associations between internalising domain and alcohol outcomes [21,24,49,54,58,[63][64][65]67,68,70]. Among the 13 studies that explored sex differences, three articles found significant sex differences [60,66,78], while the remaining ten articles reported no sex differences [46,48,51,53,55,57,62,72,77,91]. More studies are needed to draw conclusion on the potential influence of country and cohort on the association between internalising domain and alcohol use behaviours.
Results synthesised with subtype of alcohol use behaviours among high-quality studies are presented in S4 Table. There are some discrepancies with our main results: the trend that the proportion of positive associations between EXT and alcohol use increases with the severity of the outcome became less obvious (alcohol consumption: 50%; heavy/problematic drinking: 60%; alcohol use disorder: 37.5%); only negative associations were found between INT and alcohol consumption and only positive associations were detected between INT and more severe alcohol outcomes (heavy/problematic drinking, alcohol use disorder); no significant association in either direction was found between depression and alcohol use disorder; only two high-quality studies examined the association between anxiety and AUD, and one of them reported negative association.
The analysis done by extracting one association item from studies using the same dataset did not change our conclusion drawn from our main results.

Discussion
This systematic review investigated the association between early life mental health and alcohol use behaviours in adulthood. The evidence indicates positive associations between EXT and later alcohol use behaviours, but this association tends to vary with subtypes of alcohol use behaviours, with more severe outcomes being more consistently linked with EXT. EXT measured during early adolescence and adolescence appears to be more sensitive compared to that in childhood. The association between the internalising domain and alcohol use behaviours is inconclusive with both positive and negative associations presented for the same subtype of alcohol use behaviours.

Association between externalising problems and alcohol use behaviours
Our review points to a positive association between early life EXT and various alcohol use behaviours with more consistent positive associations being observed as the outcome becomes more severe (from alcohol consumption to problem drinking to AUD). This trend can also be seen from the distribution of P values across subtypes in Fig 2. Publication bias may underlie this finding since papers reporting no/negative associations are less likely to be published, but this pattern can also be seen when looking at different alcohol use behaviours within one study [23,61,64,68,92].
Based on current evidence, EXT in early adolescence and adolescence seems to play a more important role than that in childhood. This seems to contradict the hypothesis of the critical period, which emphasises that aversive experiences in late childhood (age 8-11) are especially impactful on later substance use and other behavioural problems [93]. The hypothesis states that children at this stage start to form their own identity, which is the basis for later behaviours and decisions; at the same time, they start to build affiliations with their surroundings and can easily get involved with deviance-prone peers if they manifest conduct problems themselves. Results from studies which derived trajectories of EXT seem to support the notion of "cumulative continuity". The hypothesis of cumulative continuity stresses that the continuity of EXT rather than their severity matters in the development of behavioural problems later in life [25,61,94]. Future study should focus on trajectories of EXT as an exposure to better articulate the hypothesis of critical period (adolescence versus childhood) and the notion of cumulative continuity.

Association between internalising domain and alcohol use behaviours
Compared to EXT, the associations between the subtypes of internalising domain and alcohol use behaviours are less consistent. Inconsistency may arise from the co-occurrence of EXT, which were not adjusted for in more than half of the selected studies. The proportion of positive/negative association differed between tests adjusting for EXT and those not adjusting for it, especially for INT (negative association for alcohol consumption: 87.5% vs. 18.5%). More effort needs to be made to understand how INT and EXT operate in tandem in children's lives to increase or decrease the risk for alcohol use/problems in adulthood, as EXT is quite prevalent across different levels of INT [25]. For example, a positive association between depression and AUD was found only in participants with high levels of conduct problems and not in those with low and moderate conduct problems [74]. Moreover, one small sample size study found that pure EXT (without INT) had the strongest positive association with adolescent alcohol use, but this association became weaker when EXT co-occurred with INT [27]. Meanwhile, pure INT (without EXT) presented a negative association, though it was statistically non-significant [27]. Colder et al. also found that the highest probability of alcohol use was observed in those with high EXT and low INT, and a negative association between INT and alcohol use was strongest for youth with no EXT [95].
However, although studies mentioned above consistently showed an interaction between the externalising and internalising domains, opposite associations with alcohol use were detected across subtypes of internalising domain (positive between depression and alcohol use [74] but negative between INT and alcohol use [27,95]), as is indicated by our review (See . Further implication would be that heterogeneity in measurement tools/instruments may underlie these inconsistencies in the literature as well. In 36 articles, five measurement tools were used to assess EXT, whereas more than ten tools were used for the internalising domain. It may be the case that these various tools measure different aspects of INT that exhibit different associations with alcohol use behaviours. A good illustration would be the differences between general anxiety and social anxiety. Articles that used social anxiety as exposure found negative associations with later alcohol use behaviours [53,96,97], while articles measuring general anxiety but using a scale that tended to reflect symptoms of social anxiety ("too dependent on adults," "afraid of going to school," "self-conscious or easily embarrassed," "shy or timid," "keeps from getting involved with others" [74], "fearful and helpless in other's company, target of teasing, unable to defend" [72]) also reported negative associations. Thus, it could be argued that these scales measure different aspects of anxiety, and consequently, the effect of these aspects of anxiety on alcohol behaviour may differ. For example, a person who has social anxiety might be at lower risk for getting involved with alcohol because he/she may be less exposed to other adolescents who drink, or may not have the skills to obtain alcohol if he/she is below legal drinking age [53]; however, a person with other types of anxiety may have a higher risk for later alcohol use [96].
Another finding worth our attention is how the direction of the association between internalising domain and alcohol use behaviours flipped across the subtypes of alcohol outcomes, especially for INT. One possible explanation for this might be the U or J shaped association reported in cross-sectional studies [98][99][100]. Studies that reported negative association between INT and alcohol consumption were either large sample-size studies or measured alcohol consumption in mid-adulthood [21,66,68]. Under this situation, the majority of the participants would be non-drinkers or light drinkers, and negative association would be found when the relationship was modelled as linear. By comparison, for more severe outcomes, which were mainly measured at transition to adulthood [46,54,55,63,67] when alcohol use reached its peak, the results may reflect the positive association. Interestingly, a U-shaped pattern was also observed in a recent prospective study, which discovered that adolescents with more symptoms of depression were more likely to be either abstainers or to demonstrate a problematic use [49]. Researchers should take into account the potential non-linear relationship in the future.
Even though differences were observed in some studies between males and females [21,68], sex does not appear to be a substantial factor that caused the inconsistencies in our review. However, more attention should be paid to the role sex plays in the association between early mental health and later alcohol use behaviours due to the profound sex differences in the development of mental health, physiological vulnerability to alcohol, alcohol consumption patterns, and social norms and expectations about drinking [17]. No obvious country or history differences were discovered in our review after taking other factors into account. This may indicate that the association between early life mental health and alcohol use behaviours in adulthood reflect general developmental trends rather than specific historically bounded ones or culture specific ones [22]. However, studies comparing the historical differences or crosscountries comparison (especially in non-Western countries) are needed, as none of the studies included in our review tried to answer this question directly.

Other implications for future studies
Several implications for future studies emerged from our review. Future work should examine whether the association between early life EXT/INT and alcohol use in adulthood can be interpreted as causal. Although causality in observational data is not easy to infer, a range of techniques such as cross-contextual comparisons, negative controls, sensitivity analysis for unmeasured confounders, instrumental variable analysis or Mendelian Randomization [101,102], and fixed-effect models that eliminate time-invariant confounders [103] can be used for more robust causal inference. To the best of our knowledge, only two articles in this area have applied fixed-effect models [104,105]. However, their exposure and outcome were measured within the same period and the direction of the association they found could be from alcohol use behaviours to mental health problems [104,106].
Moreover, the fact that almost half of the selected articles were rated as poor quality, and the fact that many high-quality studies did not account for missing data, raise more concern. Principled techniques to deal with missing data, such as inverse probability weighting, multiple imputations, full information maximum likelihood, or even combinations of these techniques [107][108][109] have been shown to return valid estimates under the missing at random assumption [110] and should be applied more often in the future.

Strengths and limitations
This systematic review built on previous reviews that focused on alcohol use in adolescent [16] and extended AUD into adulthood. Also, we included both domains of mental health problems (EXT and INT) and subcategorised alcohol behaviours according to their level of severity, which provided new insights into these associations. Furthermore, we summarised evidence for a potential age effect and sex differences, although no conclusive findings can be drawn. Several limitations need to be considered when interpreting the results. First, based on current theory, this review focused on broad categories of mental health problems, which resulted in missing studies on Attention Deficit Hyperactivity Disorder (ADHD) and specific anxiety subtypes. Recent studies have shown that ADHD is also positively associated with later alcohol use [88,111], and it is very likely that a particular trait within the domain of the disorder is the driver for later alcohol use [112]. Future studies should compare how the associations may change when focusing on different symptoms within a certain disorder, such as aggression, impulsivity, sensation seeking under externalising domain, and social withdrawal under internalising domain. Second, though we tried alternative ways of data synthesis to avoid the risk of bias, we were not sure about the discrepancies discovered and only reported one set of the results in detail as Hussong et al. did [16]. Results using alternative data synthesis methods are attached in S3 and S4 Tables for the readers' consideration. Third, due to the large number of articles retrieved (over 17,000), only a subset of the articles was reviewed by the second author. Forth, studies were restricted to articles published in English, and as a result, results may not be generalizable to other populations and may suffer from publication bias. However, we postulate that publication bias may exaggerate the proportion of positive associations for externalising problems to a limited extend and would not affect the results for internalising problems much, as the reported associations were already quite mixed.

Conclusion
This review evaluated the evidence on the association between early life externalising/internalising problems and alcohol use behaviours in adulthood. For externalising problems, consistent positive associations were found across studies, and there tended to be more positive associations with more severe alcohol outcomes such as heavy/problematic drinking and AUD. Externalising problems in early adolescence and adolescence seem to be more strongly associated with alcohol outcomes than that in childhood. The evidence on associations between internalising problems and alcohol use behaviours is inconclusive, and the results suggested that different domains of internalising problems may differ in their associations with later alcohol use.

S1 Fig. Distribution of P-value between EXT and phenotype of alcohol use behaviours.
Each dot represents the mean of P-value for all items that measure the corresponding association. P-value was either extracted or calculated using available information, and was coded as missing when not available. Dots on the right side of zero indicate size of P-values for positive associations, and dots on the left side of zero indicate size of P-value for negative associations. Two red lines represent a threshold of 0.05 respectively. "Adjust" means that INT(EXT) was adjusted simultaneously. This figure illustration applies to S2-S4 Figs.